AI finds new drugs through incredible 100 million molecules

Uppsala University researchers develop calculation methods that significantly reduce the time and cost of drug discovery

It seems simple to find needles in haystacks when looking for new drugs. Uppsala University scientists have developed a computational method that can evaluate almost incredible potential drug molecules (1.666 million (i.e. 22 zeros)) to identify the most promising development candidates.

The study, published in Nature Communications in February this year, demonstrates how virtual fragment screens identify molecules that inhibit OGG1, an enzyme involved in DNA repair that plays a role in inflammation and cancer.

“It’s surprising that we can now design molecules and show that they actually work exactly as we want. The same strategy will work for many other proteins and diseases.” Jense Carlson, professor of computational chemistry at Uppsala University (Jens Carlsson) said.

Build drugs like puzzles

Traditional drug discovery often relies on high-throughput screening (testing hundreds of thousands of compounds of target proteins) – which is both expensive and often unsuccessful. Instead, researchers used fragment-based drug designs, from tiny molecular fragments Start and then build them into more complex drug molecules.

“It’s like doing a puzzle puzzle. We start with a puzzle and then gradually build a drug molecule by adding new works. Finally, we have a drug molecule that is perfectly suitable for the target protein,” explains Carlson.

The team first performed calculations on the OGG1 enzyme to screen 14 million similar fragments of compounds. After identifying the promising fragments, they conducted further searches among billions of easily synthesized compounds and then performed laboratory tests on the most promising candidates.

This method successfully identified molecules that inhibit OGG1 and exhibit anti-inflammatory and anti-cancer effects in cellular testing. The crystal structure of the protein fragment complex confirmed that the compound was predicted to bind to OGG1 completely according to the computational model.

From billions to four billions

Although the initial search was limited to commercially available molecules, the researchers later greatly expanded their horizons. PhD student Andreas Luttens has developed a new computer program that is able to generate and evaluate all possible molecules that may theoretically exist within certain chemical limitations.

This expands the search space to about 1 billion potential molecules, rather than the observable number of stars in the universe. The team showed that their calculation method could effectively browse this vast chemical space to identify the most promising candidates.

“With our strategy, we can search for a quarter of drug molecules very effectively. In the near future, we will be able to test all potential drug molecules in computer models, which is a breakthrough with great potential.” Carlson explain.

Advantages of virtual clip filtering

The researchers compared the fragment-based approach with traditional screening methods and found that it was more efficient. Their analysis shows that by using a fragment-based design, they can evaluate all theoretically possible fragment-like molecules and their related elaborations through two steps for only 2 billion compounds.

By contrast, random selection of 2 billion compounds from the huge lead-like chemical space will be unlikely to include any relevant candidates targeting a specific target.

The team also tested methods for three additional protein targets associated with cancer or inflammation (SMYD3, NUDT5 and PHIP) and found that in each case, they could effectively identify promising compounds for further development.

A new era of drug discovery

While computational methods can now quickly identify promising drug candidates, converting these virtual discoveries into the real world presents new challenges.

“In the future we need to develop new methods to successfully develop molecules that can be designed so quickly,” Carlson warned.

However, this approach represents a significant advance in drug discovery methods. The ability to virtual chemistry libraries and focus resources on the most promising candidates can greatly reduce the time and cost of bringing new medications to patients.

The researchers’ OGG1 inhibitors exhibited better physicochemical and pharmacokinetic properties than previously identified compounds, indicating the practical benefits of this calculation method.

The study, conducted in collaboration with Karolinska Institutet and Stockholm University, highlights show how computational methods transform drug discovery from a largely experimental process to an increasingly increasingly guided process by sophisticated algorithms and models. .

As computing power continues to increase, these methods are perfected, and the ability to effectively browse the vast landscape of possible drug molecules may lead to faster development of more effective drugs for a variety of diseases.